營造業長期高居重大職業災害發生率之首，其中「墜落與滾落」為最主要的致死原因。傳統營建防墜風險評估多仰賴靜態檢查表與單一線性分析，難以釐清複雜施工情境下多重風險因子之交互作用。為突破此侷限，本研究導入資料探勘與機器學習技術，旨在建構一套「營建工地墜落肇因分析模式」，將被動的事故統計轉化為具備主動防範價值之圖樣識別，以深入剖析致死墜落之關鍵成因與觸發路徑。
本研究共蒐集勞動部五年期間之重大職業災害致死墜落案例，建構以 4M1E（人、機、料、法、環）架構為基礎之風險特徵矩陣。為精確捕捉動態致災脈絡，本研究導入「動作特徵（Action_Type）」與「防護類型（Prot_Type）」等關鍵變數。在資料前處理階段，應用演算法有效解決職災數據中特定災害類型之類別不平衡問題。隨後採取「雙模型比較」策略，分別建構決策樹作為解釋基準，與隨機森林進行效能競賽，並選定最佳之肇因分析工具。
三年期數據之分析結果顯示，隨機森林演算法在處理非線性且高雜訊之職災數據上展現優越效能，其肇因分類之誤判率顯著低於單一決策樹，故確立以隨機森林作為核心分析方法。進一步透過變數重要性運算，萃取出決定墜落肇因之重要性關鍵因子。此發現突破了現今營建安全管理中之兩公尺防墜計畫相關規定，證實墜落事故之型態，高度取決於勞工作業當下之動態行為（如施力、重心轉移）及現場防護設施的匹配狀態。
本研究結合決策樹產出之可視化致災路徑，具體描繪出五大墜落類型（施工架、開口部、鋼構/屋頂、機械吊掛、其他）之觸發規則。於學理上驗證了隨機森林法應用於職災肇因分析之卓越性，實務上更能協助營建事業單位將安全管理重心由「靜態法規檢核」升級為「動態行為與防護監控」，並為第一線勞動檢查提供更具針對性之精準稽查策略。

The construction industry consistently records the highest incidence of major occupational accidents, with "falls and rolls" identified as the primary cause of fatalities. Traditional fall risk assessments in construction rely heavily on static checklists and single linear analyses, which struggle to clarify the interactions of multiple risk factors within complex construction scenarios. To overcome these limitations, this study introduces data mining and machine learning techniques to construct a "Construction Site Fall Cause Analysis Model." The objective is to transform passive accident statistics into proactive pattern recognition, enabling an in-depth analysis of the key causes and trigger paths of fatal falls.
Major occupational accident cases involving fatal falls over a five-year period were collected from the Ministry of Labor to construct a risk feature matrix based on the 4M1E (Man, Machine, Material, Method, Environment) framework. To accurately capture dynamic disaster contexts, key variables such as "Action_Type" and "Prot_Type" were incorporated. During the data preprocessing stage, algorithms were applied to effectively address the class imbalance problem inherent in specific accident types within occupational disaster data. Subsequently, a "dual-model comparison" strategy was adopted, where a Decision Tree was constructed as an explanatory baseline and compared against a Random Forest model to determine the optimal cause analysis tool.
Analysis of the three-year dataset indicated that the Random Forest algorithm demonstrated superior performance in processing non-linear and high-noise occupational accident data. Its misclassification rate for cause categorization was significantly lower than that of a single Decision Tree; therefore, Random Forest was established as the core analytical method. Furthermore, through variable importance calculations, key factors determining the causes of falls were extracted. These findings challenge current regulations regarding two-meter fall protection plans in construction safety management, confirming that the patterns of fall hazards are highly dependent on the dynamic behaviors of workers at the time of the operation (such as force application and shifts in the center of gravity) and the matching status of on-site protective facilities.
By integrating the visualized disaster paths generated by the Decision Tree, the trigger rules for five major fall types (scaffolding, openings, steel structures/roofs, mechanical lifting, and others) were specifically delineated. Theoretically, this study validates the excellence of the Random Forest method in analyzing the causes of occupational accidents. Practically, it assists construction enterprises in upgrading safety management from "static regulatory compliance" to "dynamic behavior and protection monitoring," while providing more targeted and precise inspection strategies for frontline labor inspectors.

包晃豪（2011）。決策樹運用於工程查核選案之研究[碩士論文]。國立臺灣科技大學營建工程系。
江旺宗（2007）。營造業墜落意外事件模式及相關勞安法規之探討[碩士論文]。國立高雄第一科技大學營建工程系。
林楨中（2024）。墜落防止技術之探討。行政院勞工委員會勞工安全衛生研究所。
施瀞淳（2024）。機器學習方法於工地危害類型之應用[碩士論文]。朝陽科技大學。
紀佳芬、楊漢聲、陳文雄、劉國青、張庭彰、丁心逸（2008）。營造業墜落重大墜落之情境分析與預防策略。勞工安全衛生研究季刊，16（4），383-400。
張世宏（2013）。臺灣營造業重大職業傷害防止對策之研究[碩士論文]。國立臺灣大學土木工程學系。
張智奇、陳婉甄（2024）。營建工程墜落關鍵危害要因分析與預防對策研究 (計畫編號：ILOSH112-S302)。勞動部勞動及職業安全衛生研究所。
陳凱緯（2015）。以決策樹與資料探勘模型改善產品升級決策-以台灣不斷電系統為例[碩士論文]。國立成功大學工程管理碩士在職專班。
陳俊廷（2018）。應用決策樹分析重大職業災害統計資料─以建築工程為例[碩士論文]。國立臺北商業大學資訊與決策科學研究所。
勞動部（2024a）。113年營造業防墜高峰會：113年上半年營造業重大職災分析及管理作為。
勞動部（2024b）。營造工程風險評估技術指引。勞動部。
黃理彥（2025）。工程管理人員對安全設備設置認知調查-以墜落為例[碩士論文]。國立雲林科技大學營建工程系。
楊漢聲（2008）。應用基因演算法與決策樹分析營造業重大墜落職災[碩士論文]。國立台灣科技大學工業管理系。
職業安全健康局（2014）。本港建造業高處墮下的死因訊個案（1999–2011）分析。職業安全健康局。
Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32.
Bunting, J., Branche, C., Trahan, C., & Goldenhar, L. (2017). A national safety stand-down to reduce construction worker falls. Journal of Safety Research, 60, 103–111.
Chi, S., & Han, S. (2013). Analyses of systems theory for construction accident prevention with specific reference to OSHA accident reports. International Journal of Project Management, 31(7), 1027–1041.
Choi, J., Gu, B., Chin, S., & Lee, J. S. (2020). Machine learning predictive model based on national data for fatal accidents of construction workers. Automation in Construction, 110, 102974.
Dogan, A., & Birant, D. (2021). Machine learning and data mining in manufacturing. Expert Systems with Applications, 166, 114060.
Dol, S., & Jawandhiya, P. (2023). Classification technique and its combination with clustering and association rule mining in educational data mining—A survey. Engineering Applications of Artificial Intelligence, 122, 106071.
Kang, K., & Ryu, H. (2019). Predicting types of occupational accidents at construction sites in Korea using random forest model. Safety Science, 120, 226–236.
Leveson, N. (2004). A new accident model for engineering safer systems. Safety Science, 42(4), 237–270.
Mistikoglu, G., Gerek, I., Erdis, E., Usmen, P., Cakan, H., & Kazan, E. (2015). Decision tree analysis of construction fall accidents involving roofers. Expert Systems with Applications, 42(4), 2256–2263.
Zermane, A., Tohir, M. Z. M., Zermane, H., Baharun, N., & Harun, Z. (2023). Predicting fatal fall from heights accidents using random forest classification and model interpretation using SHAP. Safety Science, 159, 106021.
Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). “Why should I trust you?”: Explaining the predictions of any classifier. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 1135–1144.
Robson, L. (2020). Preventing fall-from-height injuries in construction: Effectiveness of a regulatory training standard. Journal of Safety Research, 74, 271–278.
Sarkar, S., Vinay, S., Raj, R., Maiti, J., & Mitra, P. (2019). Application of optimized machine learning techniques for prediction of occupational accidents. Computers & Operations Research, 106, 210–224.
Tixier, A. J. P., Hallowell, M. R., Rajagopalan, B., & Bowman, D. (2016). Application of machine learning to construction injury prediction. Automation in Construction, 69, 102–114.